| 摘要 |
A speech identification system for a telephone network uses neural networks to distinguish between speech and noise signals by utilizing speech and noise samples. For an input signal the respective values of linear predictor coefficients, a peak value lag of an autocorrelation function of the signal, an autocorrelation function peak value of the signal divided by the signal energy, and the number of times the 0-level is exceeded during an observation period, are determined and these values are used as input vectors to the neural network. The neural network, which has previously been trained to distinguish between speech and noise signals, calculates output values for speech and noise neurons on the basis of the inputs, and decides whether the signal is speech or noise.
|
